A Multiplicative Neural Network Architecture: Locality and Regularity of Approximation

Hee-Sun Choi, Beom-Seok Han ∗ · 2026

Read A Multiplicative Neural Network Architecture: Locality and Regularity of Approximation by Hee-Sun Choi, Beom-Seok Han ∗ (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-06374.txt.

Description

Neural networks usually build their answers by adding many small contributions. Hee-Sun Choi and Beom-Seok Han study an architecture where multiplication, not addition, is the basic building block. They prove a universal approximation theorem for it: with enough units, such networks can approximate a wide class of functions.

Their analysis focuses on how errors distribute in space and how smoothness is preserved, measured in Bessel potential spaces and with Zygmund-type seminorms. Numerical tests used targets with sharp transition layers and points where smoothness drops. The multiplicative design kept its errors tightly aligned with those rough regions, and its smoothness measures converged more steadily. Architecture choice, they conclude, has concrete analytic consequences.

Catalog record

CREATORHee-Sun Choi, Beom-Seok Han ∗
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE56 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-06374

Browse the full library · About the Tower of Babel